membranes Article
Nanofiltration and Tight Ultrafiltration Membranes for Natural Organic Matter Removal—Contribution of Fouling and Concentration Polarization to Filtration Resistance Joerg Winter 1 , Benoit Barbeau 2 1 2
*
and Pierre Bérubé 1, *
Department of Civil Engineering, The University of British Columbia, 6250 Applied Science Lane, Vancouver, BC V6T1Z4, Canada; [email protected] Department of Civil, Geological and Mining Engineering, École Polytechnique de Montréal, Montréal, QC H3T 1J4, Canada; [email protected] Correspondence: [email protected]; Tel.: +1-604-822-5665
Received: 14 March 2017; Accepted: 26 June 2017; Published: 2 July 2017
Abstract: Nanofiltration (NF) and tight ultrafiltration (tight UF) membranes are a viable treatment option for high quality drinking water production from sources with high concentrations of contaminants. To date, there is limited knowledge regarding the contribution of concentration polarization (CP) and fouling to the increase in resistance during filtration of natural organic matter (NOM) with NF and tight UF. Filtration tests were conducted with NF and tight UF membranes with molecular weight cut offs (MWCOs) of 300, 2000 and 8000 Da, and model raw waters containing different constituents of NOM. When filtering model raw waters containing high concentrations of polysaccharides (i.e., higher molecular weight NOM), the increase in resistance was dominated by fouling. When filtering model raw waters containing humic substances (i.e., lower molecular weight NOM), the increase in filtration resistance was dominated by CP. The results indicate that low MWCO membranes are better suited for NOM removal, because most of the NOM in surface waters consist mainly of humic substances, which were only effectively rejected by the lower MWCO membranes. However, when humic substances are effectively rejected, CP can become extensive, leading to a significant increase in filtration resistance by the formation of a cake/gel layer at the membrane surface. For this reason, cross-flow operation, which reduces CP, is recommended. Keywords: nanofiltration; tight ultrafiltration; concentration polarization; fouling; natural organic matter
1. Introduction Over the past 20 years, membrane filtration has been increasingly implemented in water purification processes [1]. This increase has mainly been driven by the decreasing costs of membrane systems, and increasingly stringent drinking water quality regulations. In drinking water treatment applications, ultrafiltration membranes (UF) can effectively remove particulate contaminants, such as protozoa and bacteria, from raw water sources. However, effective removal of natural organic matter (NOM) and viruses is generally limited [2,3]. This is because the molecular weight cut off (MWCO) of UF membranes typically used in drinking water treatment applications is relatively large (i.e., greater than 100,000 Da). Membranes with a MWCO of less than 10,000 Da are required to effectively remove NOM and all pathogens [4–8]. The present study investigated the use of membranes with MWCOs ranging from 300 to 8000 Da for the removal of NOM. In this range, membranes are commonly referred to as nanofiltration (NF) membranes (i.e., with a typical MWCO range of 200–1000 Da) and tight UF membranes (i.e., with a typical MWCO range of 1000 to 10,000 Da). The removal of NOM from source waters is of importance in drinking water treatment because it can cause color, taste, and odor issues, Membranes 2017, 7, 34; doi:10.3390/membranes7030034
www.mdpi.com/journal/membranes
Membranes 2017, 7, 34
2 of 14
increase chlorine demand, as well as contribute to disinfection by-product formation and microbial regrowth in distribution systems [9–14]. Because NF and tight UF membranes can achieve sufficient primary disinfection (i.e., >4-log removal of all pathogens) and extensive removal of NOM, they can provide effective and comprehensive drinking water treatment in a single step. A simple, single-step approach is of particular interest for small/remote communities, because of the limited financial and technical resources generally available to implement and operate water treatment systems. The use of NF and tight UF membranes in drinking water treatment applications is still limited. This is mainly because spiral wound configurations, which generally require extensive pre-treatment and a high trans-membrane pressure, have historically been used for NF and tight UF membrane applications. In addition, fouling control measures, such as backwashing or surface scouring, cannot be applied to spiral wound configurations. Recent developments in NF and tight UF membrane configurations, such as hollow fiber configurations, promise to address some of these limitations [15–18]. Frank et al. have also reported that a substantially higher permeability can be maintained using hollow fiber configurations [16]. However, other studies have reported that hollow fiber configurations are less effective in mitigating concentration polarization (CP) than spiral wound configurations [19,20]. Because CP and fouling are still major challenges in the application of NF and tight UF membranes for drinking water treatment, insight into CP and fouling is necessary to develop recommendations for the design and operation of NF and tight UF membranes, and to enable greater adoption of this advanced treatment technology. Fouling occurs as material that is retained at the membrane surface or inside the membrane pores, accumulates, and increases resistance to the permeate flow. Material accumulates when the permeation drag—the force that transports potential foulants towards the membrane—is greater than forces acting in the opposite direction (i.e., away from the membrane) [21,22]. NOM is generally considered to be a main contributor to membrane fouling in drinking water treatment applications [4,23–27]. The extent of fouling is not necessarily proportional to the total amount of NOM retained, but is rather governed by the retention of specific NOM fractions [28–31]. The NOM fractions that have been reported to be mainly relevant to membrane fouling are biopolymers and humic substances, which are mainly of aquatic and terrestrial origin, respectively [32–34]. For UF membranes, the biopolymer fraction of NOM generally contributes the most to fouling [35–38]. Fouling due to humic substances is generally not as extensive, although it is more difficult to control hydraulically (e.g., by backwashing) [25,35]. The extent and the hydraulic reversibility of UF fouling can also be affected by other constituents in raw waters, notably calcium [25,35,39]. Calcium can form bridges between NOM molecules, between the membrane surface and NOM, as well as contribute to aggregation of NOM by charge destabilization [25,35,39]. Biopolymers have also been reported to substantially contribute to fouling of NF and tight UF membranes [28,38,40]. Humic substances, although effectively rejected, have not been reported to substantially contribute to fouling of NF membranes [41]. The limited contribution of humic substances to fouling has been attributed to charge repulsion effects, which enhance the back transport of humic substances in cross-flow systems [41,42]. The extent to which electrostatic repulsion contributes to fouling control has been demonstrated to increase with the ratio of cross-flow velocity to permeate flux [39]. Therefore, the effect of electrostatic repulsion on fouling due to humic substances, is likely to be substantially less pronounced in dead-end systems. The reported effects of calcium on fouling in NF and tight UF membranes systems have been inconsistent, and likely depend on the concentration in the solution being filtered [29,39,43,44]. CP, resulting from the accumulation of dissolved material rejected by the membrane, can also increase the resistance to permeate flow [45–47]. Unlike the situation with fouling, material does not deposit on the membrane surface or inside the membrane pores, but accumulates in proximity to the membrane surface. CP is characterized by an equilibrium between the convective transport of material towards the membrane, and the diffusive transport of retained material away from the membrane; a steady state that is expected to develop within the first minutes of filtration [29,45,48]. CP can affect the resistance to permeate flow by increasing the viscosity, the osmotic pressure of the solution being
Membranes 2017, 7, 34
3 of 14
filtered, and the back-diffusion of retained solutes [45]. If CP effects become extensive, solutes at the membrane surface can reach a critical concentration (i.e., solubility limit), beyond which they form a cake/gel layer which fouls the membrane surface [26,46–48]. In literature, this critical concentration is also referred to as the “gel concentration” [46,47]. If a material reaches the critical concentration, the back-diffusion of material is limited by the formation of a cake/gel layer. Under such conditions, the convective transport towards the membrane becomes greater than the diffusive transport away from the membrane, and fouling occurs, which increases the resistance to permeate flow throughout the filtration phase. According to the Stokes–Einstein equation, the diffusivity and the diffusive transport of material depends on its size. NOM ranges in size from a few hundred Daltons (low molecular weight acids and neutrals) to over 20,000 Da (biopolymers) [39,49–51]. The size of NOM also generally increases with ionic strength, and decreases with pH of a solution, due to changes in the shape of NOM from uncoiled to coiled [52]. CP due to the retention of salts in Reverse Osmosis (RO) and NF systems has been extensively investigated [19,53–56]. However, only a few studies have investigated CP due to the retention of NOM in NF membranes. These studies have indicated that NOM can result in CP, and therefore, affect system performance [26,57]. Also, comprehensive knowledge on the effect of (i) the type of NOM (e.g., polysaccharides and humic substances), and (ii) the membrane MWCO on the contribution of CP to the total increase in filtration resistance, is not currently available. This is a knowledge gap that limits the ability to develop recommendations for the optimal design and operation of NF and tight membrane systems for drinking water treatment. The objective of the present study was to quantify the contribution of fouling and CP to the total increase in resistance during filtration of model raw waters containing polysaccharides, humic substances, and a mixture of both (as typical constituents of NOM in natural surface waters), and to identify conditions under which an extensive increase in filtration resistance occurs. Since fouling and CP are not only expected to depend on the raw water characteristics, but also on the membrane’s selectivity, the impact of the MWCO of a membrane on CP and fouling was also investigated. As previously discussed, CP is expected to develop rapidly within the first minutes of filtration [39,45]. However, this has never been investigated for NF and tight UF membranes in drinking water applications. 2. Material and Methods 2.1. Experimental Approach Model raw waters of various compositions were considered. The biopolymer and humic material content of the raw waters were modeled using polysaccharide alginate derived from brown algae (Sodium Alginate, Sigma Aldrich, Oakville, ON , Canada) and Suwannee River NOM (SRNOM) 2R101N (International Humic Substances Society, St. Paul, MN, USA), respectively. Model raw waters with three different NOM compositions were considered: (i) alginate, (ii) SRNOM, and (iii) mixtures of SRNOM and alginate with a carbon mass ratio of 4:1. The ratio of 4:1 in the mixtures of SRNOM and alginate was selected based on humic substances to biopolymers ratios obtained from size exclusion chromatography (SEC) analyses of local surface water (Jericho Pond, Vancouver, BC, Canada) and is also consistent with the ratios reported in literature, which indicates that humic substances account for 70–80% of the NOM in natural waters [32]. All three model raw water compositions were tested at two dissolved organic carbon (DOC). concentrations: 5 mg/L, and 10 mg/L. To prepare the model raw waters, NOM surrogates (i.e., the polysaccharide alginate and/or SRNOM) were dissolved in Milli-Q laboratory water (MilliporeSigma, Temecula, CA, USA). Sodium bicarbonate at a concentration of 1 mM was added as a buffer. This concentration of bicarbonate also corresponds to an alkalinity that is within a range typical of that present in natural source waters [58]. The retention of alkalinity during filtration ranged from approximately 2% to 40% (depending on the membrane MWCO), and the impact of alkalinity rejection on CP and resistance during filtration was calculated [28], and found to be negligible relative to the overall increases in
Membranes 2017, 7, 34
4 of 14
Membranes 2017, 7, 34 membrane MWCO), and the
of 14 impact of alkalinity rejection on CP and resistance during filtration4 was calculated [28], and found to be negligible relative to the overall increases in resistance observed in the presentobserved study (results shown), and (results hence, isnot notshown), further discussed. pH was thendiscussed. adjusted resistance in thenot present study and hence,The is not further to 7.0 (± 0.1), using sodium hydroxide or hydrochloric acid, as needed. Prior to all filtration all The pH was then adjusted to 7.0 (± 0.1), using sodium hydroxide or hydrochloric acid, as tests, needed. modeltoraw water solutions through a 0.45were μm filtered nitrocellulose laboratory Prior all filtration tests, all were modelfiltered raw water solutions throughfilter. a 0.45Milli-Q µm nitrocellulose water was used for clean water flux tests priorand post-filtration of model raw waters. pH of filter. Milli-Q laboratory water was used for clean water flux tests prior- and post-filtrationThe of model the Milli-Q laboratory water used for clean water flux tests was also adjusted to 7.0 (±0.1), using raw waters. The pH of the Milli-Q laboratory water used for clean water flux tests was also adjusted to sodium hydroxide. 7.0 (±0.1), using sodium hydroxide. The experimental experimental setup, setup, illustrated illustrated in in Figure Figure 1, 1, consisted consisted of of feed feed vessels, vessels, membrane membrane cells, cells, bleed bleed The lines and and permeate permeate collection collection systems. feed vessels vessels were were used, used, one one containing containing clean clean water water and and lines systems. Two Two feed the other containing a model raw water. The pressure applied to both vessels was equal, enabling the the other containing a model raw water. The pressure applied to both vessels was equal, enabling feed to the membrane to be switched between model raw water and clean water without causing any the feed to the membrane to be switched between model raw water and clean water without causing pressure variation in theinmembrane cells. cells. ThreeThree CF042CF042 membrane cells (Sterlitech, Seattle,Seattle, WA, USA) any pressure variation the membrane membrane cells (Sterlitech, WA, were used in parallel. These accommodated flat sheet membrane coupons, 39.2 mm wide and 85.5 USA) were used in parallel. These accommodated flat sheet membrane coupons, 39.2 mm wide and mm mm long.long. Flat Flat sheet, thin thin filmfilm composite polyamide (PA)(PA) membranes with nominal MWCOs of 85.5 sheet, composite polyamide membranes with nominal MWCOs approximately 300300 DaDa (DK series, GE of approximately (DK series,GE GEOsmonics, Osmonics,Trevose, Trevose,PA, PA,USA), USA), 2000 2000 Da Da (GH (GH series, series, GE Osmonics) and 8000 Da (GM series, GE Osmonics) were considered as a representative range for NF– Osmonics) and 8000 Da (GM series, GE Osmonics) were considered as a representative range for tight UF range membranes. The intrinsic membrane resistances for the for 300 the Da, 300 2000Da, Da2000 and 8000 Da NF–tight UF range membranes. The intrinsic membrane resistances Da and 13 13 −1 13 13 −1 13 MWCO were 9.0 ×were 10 ±9.0 1.1××10 1013 m , 8.1××10 1013 m ± 1.1 10 ×m 2.2 × × 10 2.0−1× and 1012 −1 , ×8.1 ± 1.1 ± 1.1 8000 Da membranes MWCO membranes 1013and 1013± m m−1,×respectively. pressure applied to the vessels wasapplied kept constant at 40.0 ±was 1.0 2.2 1013 ± 2.0 ×The 1012transmembrane m−1 , respectively. The transmembrane pressure to the vessels psi (i.e., 2.8 ± 0.1 The conducted at room (23 ± 1 °C).atThe membrane cells kept constant at bar). 40.0 ± 1.0tests psi were (i.e., 2.8 ± 0.1 bar). The temperature tests were conducted room temperature were operated in a pseudo dead-end mode with a continuous bleed that introduced a very low and ◦ (23 ± 1 C). The membrane cells were operated in a pseudo dead-end mode with a continuous bleed constant cross-flow of less than 0.006 m/min along the membrane surface. The bleed line allowed the that introduced a very low and constant cross-flow of less than 0.006 m/min along the membrane liquid to The be purged fromallowed the membrane cellto when the feedfrom was the switched, whilecell stillwhen providing pseudo surface. bleed line the liquid be purged membrane the feed was dead-end operation. Based on preliminary tests, this cross-flow velocity had a negligible effect on switched, while still providing pseudo dead-end operation. Based on preliminary tests, this cross-flow fouling and concentration polarization (results not presented). All filtration tests were conducted in velocity had a negligible effect on fouling and concentration polarization (results not presented). triplicate in three phases: All filtration tests sequential were conducted in triplicate in three sequential phases:
1. 1. 2. 2. 3. 3.
Pre-clean Pre-clean water water filtration filtration phase phase (for (for aa minimum minimum of of 33 h). h). Filtration phase with model raw waters (for approximately Filtration phase with model raw waters (for approximately 33 h). h). Post-clean water filtration phase (for a minimum of 45 min). Post-clean water filtration phase (for a minimum of 45 min). The filtration tests tests were were conducted conducted in in these these three three sequential sequential phases phases to to quantify quantify the the contribution contribution The filtration of CP CP and and fouling fouling to to the the increase increase in in the the resistance resistance to as discussed discussed in in more more detail detail in in of to permeate permeate flow, flow, as Section 3.2. Section 3.2.
Figure 1. 1. Experimental setup. Figure Experimental setup.
Immediately prior prior to to the the start the drain drain valve valve was was Immediately start of of the the filtration filtration phase phase with with model model raw raw water, water, the opened and and the the content content of the flow flow cell cell was was purged, purged, ensuring ensuring that that the the concentration concentration of of NOM NOM in in the the opened of the
Membranes 2017, 7, 34
5 of 14
flow cell at the start of the filtration of model raw water, was equivalent to that of the model raw water. The drain was then partially closed to generate a constant low cross-flow velocity of 0.006 m/min. 2.2. Data Evaluation and Sample Analyses The rate of fouling was quantified based on fouling coefficients obtained from standard filtration laws fitted to the experimental data collected during the filtration of model raw water [59]. Based on the minimum R2 value for all conditions investigated in the present study, the permeate flux could be best modeled by assuming that fouling was predominantly due to the formation of a cake layer on the membrane surface (see Equation (1); results not presented). Therefore, the results are presented in terms of a cake fouling coefficient (kc ) (see Table 1). J=
1 + kc ×
J00 µ ∆P
h
× V×
J00
m3 m − 2 s − 1
i
(1)
In Equation (1), J represents the permeate flux [m3 m−2 s−1 ], V the specific volume of water filtered [m3 m−2 ], µ the dynamic viscosity [Pa s] and ∆P the trans-membrane pressure [Pa]. The initial flux J00 was defined as the measured clean water flux (J0 ), minus any rapid decrease observed in the flux within the first few minutes of a filtration test. Because CP is expected to develop within the first minutes of filtration, the difference between J0 and J00 was assumed to be due to CP [39,45]. The estimated cake fouling coefficient kc [m−1 m−3 m2 ] quantifies the increase in resistance [m−1 ] per specific volume filtered [m3 m−2 ]. Size exclusion chromatography (SEC) was performed using High Performance Liquid Chromatography (HPLC) (Perkin Elmer, Burnaby, BC, Canada) with DOC detection for the analysis of NOM in the feed and permeate. The method used was adopted from Huber et al. [51]. A TSK HW-50S column (Tosoh, Tokyo, Japan) was used as the stationary phase, and a phosphate buffer (2.5 g/L KH2 PO4 + 1.5 g/L Na2 HPO4 ·H2 O) was used as the mobile phase. The sample injection volume and flow rate were 1 mL and 1 mL/min, respectively. A GE Sievers 900 Turbo Portable TOC Analyzer (GE Sievers, Boulder, CO, USA) with a sampling rate of 4 s and a detection range of 0.2 mg C/L to 10 mg C/L was used as the DOC detector. 3. Results 3.1. NOM Rejection The amount of material rejected by the membranes was defined as the difference between material present in the feed (i.e., the model raw water) and material present in the permeate. As illustrated in Figure 2, the membranes with MWCOs of 300 Da and 2000 Da could effectively reject all of the alginate present in model raw waters, and the membrane with a MWCO of 8000 Da could reject most of the alginate present in model raw waters. This was expected because the molecular weight of alginate is greater than 10,000 Da [39,51] and the MWCOs of all the membranes considered are lower than 10,000 Da (i.e., ranging from 300 Da to 8000 Da). Only a small amount of the larger constituents in SRNOM was rejected by the membrane with a MWCO of 8000 Da. The extent of the rejection increased as the MWCO of the membranes decreased, with all of the organic material being rejected by the membrane with a MWCO of 300 Da. Again, this was expected, because most of the constituents of SRNOM have a molecular weight in between 300 Da and 8000 Da [49,50]. Note that NOM in most surface waters consists mainly of humic substances (i.e., lower molecular weight NOM) [32]. Because effective NOM removal is one of the main motivations behind implementing NF and tight UF membranes in drinking water treatment, membranes with lower MWCOs (i.e.,
Nanofiltration and Tight Ultrafiltration Membranes for Natural Organic Matter Removal—Contribution of Fouling and Concentration Polarization to Filtration Resistance Joerg Winter 1 , Benoit Barbeau 2 1 2
*
and Pierre Bérubé 1, *
Department of Civil Engineering, The University of British Columbia, 6250 Applied Science Lane, Vancouver, BC V6T1Z4, Canada; [email protected] Department of Civil, Geological and Mining Engineering, École Polytechnique de Montréal, Montréal, QC H3T 1J4, Canada; [email protected] Correspondence: [email protected]; Tel.: +1-604-822-5665
Received: 14 March 2017; Accepted: 26 June 2017; Published: 2 July 2017
Abstract: Nanofiltration (NF) and tight ultrafiltration (tight UF) membranes are a viable treatment option for high quality drinking water production from sources with high concentrations of contaminants. To date, there is limited knowledge regarding the contribution of concentration polarization (CP) and fouling to the increase in resistance during filtration of natural organic matter (NOM) with NF and tight UF. Filtration tests were conducted with NF and tight UF membranes with molecular weight cut offs (MWCOs) of 300, 2000 and 8000 Da, and model raw waters containing different constituents of NOM. When filtering model raw waters containing high concentrations of polysaccharides (i.e., higher molecular weight NOM), the increase in resistance was dominated by fouling. When filtering model raw waters containing humic substances (i.e., lower molecular weight NOM), the increase in filtration resistance was dominated by CP. The results indicate that low MWCO membranes are better suited for NOM removal, because most of the NOM in surface waters consist mainly of humic substances, which were only effectively rejected by the lower MWCO membranes. However, when humic substances are effectively rejected, CP can become extensive, leading to a significant increase in filtration resistance by the formation of a cake/gel layer at the membrane surface. For this reason, cross-flow operation, which reduces CP, is recommended. Keywords: nanofiltration; tight ultrafiltration; concentration polarization; fouling; natural organic matter
1. Introduction Over the past 20 years, membrane filtration has been increasingly implemented in water purification processes [1]. This increase has mainly been driven by the decreasing costs of membrane systems, and increasingly stringent drinking water quality regulations. In drinking water treatment applications, ultrafiltration membranes (UF) can effectively remove particulate contaminants, such as protozoa and bacteria, from raw water sources. However, effective removal of natural organic matter (NOM) and viruses is generally limited [2,3]. This is because the molecular weight cut off (MWCO) of UF membranes typically used in drinking water treatment applications is relatively large (i.e., greater than 100,000 Da). Membranes with a MWCO of less than 10,000 Da are required to effectively remove NOM and all pathogens [4–8]. The present study investigated the use of membranes with MWCOs ranging from 300 to 8000 Da for the removal of NOM. In this range, membranes are commonly referred to as nanofiltration (NF) membranes (i.e., with a typical MWCO range of 200–1000 Da) and tight UF membranes (i.e., with a typical MWCO range of 1000 to 10,000 Da). The removal of NOM from source waters is of importance in drinking water treatment because it can cause color, taste, and odor issues, Membranes 2017, 7, 34; doi:10.3390/membranes7030034
www.mdpi.com/journal/membranes
Membranes 2017, 7, 34
2 of 14
increase chlorine demand, as well as contribute to disinfection by-product formation and microbial regrowth in distribution systems [9–14]. Because NF and tight UF membranes can achieve sufficient primary disinfection (i.e., >4-log removal of all pathogens) and extensive removal of NOM, they can provide effective and comprehensive drinking water treatment in a single step. A simple, single-step approach is of particular interest for small/remote communities, because of the limited financial and technical resources generally available to implement and operate water treatment systems. The use of NF and tight UF membranes in drinking water treatment applications is still limited. This is mainly because spiral wound configurations, which generally require extensive pre-treatment and a high trans-membrane pressure, have historically been used for NF and tight UF membrane applications. In addition, fouling control measures, such as backwashing or surface scouring, cannot be applied to spiral wound configurations. Recent developments in NF and tight UF membrane configurations, such as hollow fiber configurations, promise to address some of these limitations [15–18]. Frank et al. have also reported that a substantially higher permeability can be maintained using hollow fiber configurations [16]. However, other studies have reported that hollow fiber configurations are less effective in mitigating concentration polarization (CP) than spiral wound configurations [19,20]. Because CP and fouling are still major challenges in the application of NF and tight UF membranes for drinking water treatment, insight into CP and fouling is necessary to develop recommendations for the design and operation of NF and tight UF membranes, and to enable greater adoption of this advanced treatment technology. Fouling occurs as material that is retained at the membrane surface or inside the membrane pores, accumulates, and increases resistance to the permeate flow. Material accumulates when the permeation drag—the force that transports potential foulants towards the membrane—is greater than forces acting in the opposite direction (i.e., away from the membrane) [21,22]. NOM is generally considered to be a main contributor to membrane fouling in drinking water treatment applications [4,23–27]. The extent of fouling is not necessarily proportional to the total amount of NOM retained, but is rather governed by the retention of specific NOM fractions [28–31]. The NOM fractions that have been reported to be mainly relevant to membrane fouling are biopolymers and humic substances, which are mainly of aquatic and terrestrial origin, respectively [32–34]. For UF membranes, the biopolymer fraction of NOM generally contributes the most to fouling [35–38]. Fouling due to humic substances is generally not as extensive, although it is more difficult to control hydraulically (e.g., by backwashing) [25,35]. The extent and the hydraulic reversibility of UF fouling can also be affected by other constituents in raw waters, notably calcium [25,35,39]. Calcium can form bridges between NOM molecules, between the membrane surface and NOM, as well as contribute to aggregation of NOM by charge destabilization [25,35,39]. Biopolymers have also been reported to substantially contribute to fouling of NF and tight UF membranes [28,38,40]. Humic substances, although effectively rejected, have not been reported to substantially contribute to fouling of NF membranes [41]. The limited contribution of humic substances to fouling has been attributed to charge repulsion effects, which enhance the back transport of humic substances in cross-flow systems [41,42]. The extent to which electrostatic repulsion contributes to fouling control has been demonstrated to increase with the ratio of cross-flow velocity to permeate flux [39]. Therefore, the effect of electrostatic repulsion on fouling due to humic substances, is likely to be substantially less pronounced in dead-end systems. The reported effects of calcium on fouling in NF and tight UF membranes systems have been inconsistent, and likely depend on the concentration in the solution being filtered [29,39,43,44]. CP, resulting from the accumulation of dissolved material rejected by the membrane, can also increase the resistance to permeate flow [45–47]. Unlike the situation with fouling, material does not deposit on the membrane surface or inside the membrane pores, but accumulates in proximity to the membrane surface. CP is characterized by an equilibrium between the convective transport of material towards the membrane, and the diffusive transport of retained material away from the membrane; a steady state that is expected to develop within the first minutes of filtration [29,45,48]. CP can affect the resistance to permeate flow by increasing the viscosity, the osmotic pressure of the solution being
Membranes 2017, 7, 34
3 of 14
filtered, and the back-diffusion of retained solutes [45]. If CP effects become extensive, solutes at the membrane surface can reach a critical concentration (i.e., solubility limit), beyond which they form a cake/gel layer which fouls the membrane surface [26,46–48]. In literature, this critical concentration is also referred to as the “gel concentration” [46,47]. If a material reaches the critical concentration, the back-diffusion of material is limited by the formation of a cake/gel layer. Under such conditions, the convective transport towards the membrane becomes greater than the diffusive transport away from the membrane, and fouling occurs, which increases the resistance to permeate flow throughout the filtration phase. According to the Stokes–Einstein equation, the diffusivity and the diffusive transport of material depends on its size. NOM ranges in size from a few hundred Daltons (low molecular weight acids and neutrals) to over 20,000 Da (biopolymers) [39,49–51]. The size of NOM also generally increases with ionic strength, and decreases with pH of a solution, due to changes in the shape of NOM from uncoiled to coiled [52]. CP due to the retention of salts in Reverse Osmosis (RO) and NF systems has been extensively investigated [19,53–56]. However, only a few studies have investigated CP due to the retention of NOM in NF membranes. These studies have indicated that NOM can result in CP, and therefore, affect system performance [26,57]. Also, comprehensive knowledge on the effect of (i) the type of NOM (e.g., polysaccharides and humic substances), and (ii) the membrane MWCO on the contribution of CP to the total increase in filtration resistance, is not currently available. This is a knowledge gap that limits the ability to develop recommendations for the optimal design and operation of NF and tight membrane systems for drinking water treatment. The objective of the present study was to quantify the contribution of fouling and CP to the total increase in resistance during filtration of model raw waters containing polysaccharides, humic substances, and a mixture of both (as typical constituents of NOM in natural surface waters), and to identify conditions under which an extensive increase in filtration resistance occurs. Since fouling and CP are not only expected to depend on the raw water characteristics, but also on the membrane’s selectivity, the impact of the MWCO of a membrane on CP and fouling was also investigated. As previously discussed, CP is expected to develop rapidly within the first minutes of filtration [39,45]. However, this has never been investigated for NF and tight UF membranes in drinking water applications. 2. Material and Methods 2.1. Experimental Approach Model raw waters of various compositions were considered. The biopolymer and humic material content of the raw waters were modeled using polysaccharide alginate derived from brown algae (Sodium Alginate, Sigma Aldrich, Oakville, ON , Canada) and Suwannee River NOM (SRNOM) 2R101N (International Humic Substances Society, St. Paul, MN, USA), respectively. Model raw waters with three different NOM compositions were considered: (i) alginate, (ii) SRNOM, and (iii) mixtures of SRNOM and alginate with a carbon mass ratio of 4:1. The ratio of 4:1 in the mixtures of SRNOM and alginate was selected based on humic substances to biopolymers ratios obtained from size exclusion chromatography (SEC) analyses of local surface water (Jericho Pond, Vancouver, BC, Canada) and is also consistent with the ratios reported in literature, which indicates that humic substances account for 70–80% of the NOM in natural waters [32]. All three model raw water compositions were tested at two dissolved organic carbon (DOC). concentrations: 5 mg/L, and 10 mg/L. To prepare the model raw waters, NOM surrogates (i.e., the polysaccharide alginate and/or SRNOM) were dissolved in Milli-Q laboratory water (MilliporeSigma, Temecula, CA, USA). Sodium bicarbonate at a concentration of 1 mM was added as a buffer. This concentration of bicarbonate also corresponds to an alkalinity that is within a range typical of that present in natural source waters [58]. The retention of alkalinity during filtration ranged from approximately 2% to 40% (depending on the membrane MWCO), and the impact of alkalinity rejection on CP and resistance during filtration was calculated [28], and found to be negligible relative to the overall increases in
Membranes 2017, 7, 34
4 of 14
Membranes 2017, 7, 34 membrane MWCO), and the
of 14 impact of alkalinity rejection on CP and resistance during filtration4 was calculated [28], and found to be negligible relative to the overall increases in resistance observed in the presentobserved study (results shown), and (results hence, isnot notshown), further discussed. pH was thendiscussed. adjusted resistance in thenot present study and hence,The is not further to 7.0 (± 0.1), using sodium hydroxide or hydrochloric acid, as needed. Prior to all filtration all The pH was then adjusted to 7.0 (± 0.1), using sodium hydroxide or hydrochloric acid, as tests, needed. modeltoraw water solutions through a 0.45were μm filtered nitrocellulose laboratory Prior all filtration tests, all were modelfiltered raw water solutions throughfilter. a 0.45Milli-Q µm nitrocellulose water was used for clean water flux tests priorand post-filtration of model raw waters. pH of filter. Milli-Q laboratory water was used for clean water flux tests prior- and post-filtrationThe of model the Milli-Q laboratory water used for clean water flux tests was also adjusted to 7.0 (±0.1), using raw waters. The pH of the Milli-Q laboratory water used for clean water flux tests was also adjusted to sodium hydroxide. 7.0 (±0.1), using sodium hydroxide. The experimental experimental setup, setup, illustrated illustrated in in Figure Figure 1, 1, consisted consisted of of feed feed vessels, vessels, membrane membrane cells, cells, bleed bleed The lines and and permeate permeate collection collection systems. feed vessels vessels were were used, used, one one containing containing clean clean water water and and lines systems. Two Two feed the other containing a model raw water. The pressure applied to both vessels was equal, enabling the the other containing a model raw water. The pressure applied to both vessels was equal, enabling feed to the membrane to be switched between model raw water and clean water without causing any the feed to the membrane to be switched between model raw water and clean water without causing pressure variation in theinmembrane cells. cells. ThreeThree CF042CF042 membrane cells (Sterlitech, Seattle,Seattle, WA, USA) any pressure variation the membrane membrane cells (Sterlitech, WA, were used in parallel. These accommodated flat sheet membrane coupons, 39.2 mm wide and 85.5 USA) were used in parallel. These accommodated flat sheet membrane coupons, 39.2 mm wide and mm mm long.long. Flat Flat sheet, thin thin filmfilm composite polyamide (PA)(PA) membranes with nominal MWCOs of 85.5 sheet, composite polyamide membranes with nominal MWCOs approximately 300300 DaDa (DK series, GE of approximately (DK series,GE GEOsmonics, Osmonics,Trevose, Trevose,PA, PA,USA), USA), 2000 2000 Da Da (GH (GH series, series, GE Osmonics) and 8000 Da (GM series, GE Osmonics) were considered as a representative range for NF– Osmonics) and 8000 Da (GM series, GE Osmonics) were considered as a representative range for tight UF range membranes. The intrinsic membrane resistances for the for 300 the Da, 300 2000Da, Da2000 and 8000 Da NF–tight UF range membranes. The intrinsic membrane resistances Da and 13 13 −1 13 13 −1 13 MWCO were 9.0 ×were 10 ±9.0 1.1××10 1013 m , 8.1××10 1013 m ± 1.1 10 ×m 2.2 × × 10 2.0−1× and 1012 −1 , ×8.1 ± 1.1 ± 1.1 8000 Da membranes MWCO membranes 1013and 1013± m m−1,×respectively. pressure applied to the vessels wasapplied kept constant at 40.0 ±was 1.0 2.2 1013 ± 2.0 ×The 1012transmembrane m−1 , respectively. The transmembrane pressure to the vessels psi (i.e., 2.8 ± 0.1 The conducted at room (23 ± 1 °C).atThe membrane cells kept constant at bar). 40.0 ± 1.0tests psi were (i.e., 2.8 ± 0.1 bar). The temperature tests were conducted room temperature were operated in a pseudo dead-end mode with a continuous bleed that introduced a very low and ◦ (23 ± 1 C). The membrane cells were operated in a pseudo dead-end mode with a continuous bleed constant cross-flow of less than 0.006 m/min along the membrane surface. The bleed line allowed the that introduced a very low and constant cross-flow of less than 0.006 m/min along the membrane liquid to The be purged fromallowed the membrane cellto when the feedfrom was the switched, whilecell stillwhen providing pseudo surface. bleed line the liquid be purged membrane the feed was dead-end operation. Based on preliminary tests, this cross-flow velocity had a negligible effect on switched, while still providing pseudo dead-end operation. Based on preliminary tests, this cross-flow fouling and concentration polarization (results not presented). All filtration tests were conducted in velocity had a negligible effect on fouling and concentration polarization (results not presented). triplicate in three phases: All filtration tests sequential were conducted in triplicate in three sequential phases:
1. 1. 2. 2. 3. 3.
Pre-clean Pre-clean water water filtration filtration phase phase (for (for aa minimum minimum of of 33 h). h). Filtration phase with model raw waters (for approximately Filtration phase with model raw waters (for approximately 33 h). h). Post-clean water filtration phase (for a minimum of 45 min). Post-clean water filtration phase (for a minimum of 45 min). The filtration tests tests were were conducted conducted in in these these three three sequential sequential phases phases to to quantify quantify the the contribution contribution The filtration of CP CP and and fouling fouling to to the the increase increase in in the the resistance resistance to as discussed discussed in in more more detail detail in in of to permeate permeate flow, flow, as Section 3.2. Section 3.2.
Figure 1. 1. Experimental setup. Figure Experimental setup.
Immediately prior prior to to the the start the drain drain valve valve was was Immediately start of of the the filtration filtration phase phase with with model model raw raw water, water, the opened and and the the content content of the flow flow cell cell was was purged, purged, ensuring ensuring that that the the concentration concentration of of NOM NOM in in the the opened of the
Membranes 2017, 7, 34
5 of 14
flow cell at the start of the filtration of model raw water, was equivalent to that of the model raw water. The drain was then partially closed to generate a constant low cross-flow velocity of 0.006 m/min. 2.2. Data Evaluation and Sample Analyses The rate of fouling was quantified based on fouling coefficients obtained from standard filtration laws fitted to the experimental data collected during the filtration of model raw water [59]. Based on the minimum R2 value for all conditions investigated in the present study, the permeate flux could be best modeled by assuming that fouling was predominantly due to the formation of a cake layer on the membrane surface (see Equation (1); results not presented). Therefore, the results are presented in terms of a cake fouling coefficient (kc ) (see Table 1). J=
1 + kc ×
J00 µ ∆P
h
× V×
J00
m3 m − 2 s − 1
i
(1)
In Equation (1), J represents the permeate flux [m3 m−2 s−1 ], V the specific volume of water filtered [m3 m−2 ], µ the dynamic viscosity [Pa s] and ∆P the trans-membrane pressure [Pa]. The initial flux J00 was defined as the measured clean water flux (J0 ), minus any rapid decrease observed in the flux within the first few minutes of a filtration test. Because CP is expected to develop within the first minutes of filtration, the difference between J0 and J00 was assumed to be due to CP [39,45]. The estimated cake fouling coefficient kc [m−1 m−3 m2 ] quantifies the increase in resistance [m−1 ] per specific volume filtered [m3 m−2 ]. Size exclusion chromatography (SEC) was performed using High Performance Liquid Chromatography (HPLC) (Perkin Elmer, Burnaby, BC, Canada) with DOC detection for the analysis of NOM in the feed and permeate. The method used was adopted from Huber et al. [51]. A TSK HW-50S column (Tosoh, Tokyo, Japan) was used as the stationary phase, and a phosphate buffer (2.5 g/L KH2 PO4 + 1.5 g/L Na2 HPO4 ·H2 O) was used as the mobile phase. The sample injection volume and flow rate were 1 mL and 1 mL/min, respectively. A GE Sievers 900 Turbo Portable TOC Analyzer (GE Sievers, Boulder, CO, USA) with a sampling rate of 4 s and a detection range of 0.2 mg C/L to 10 mg C/L was used as the DOC detector. 3. Results 3.1. NOM Rejection The amount of material rejected by the membranes was defined as the difference between material present in the feed (i.e., the model raw water) and material present in the permeate. As illustrated in Figure 2, the membranes with MWCOs of 300 Da and 2000 Da could effectively reject all of the alginate present in model raw waters, and the membrane with a MWCO of 8000 Da could reject most of the alginate present in model raw waters. This was expected because the molecular weight of alginate is greater than 10,000 Da [39,51] and the MWCOs of all the membranes considered are lower than 10,000 Da (i.e., ranging from 300 Da to 8000 Da). Only a small amount of the larger constituents in SRNOM was rejected by the membrane with a MWCO of 8000 Da. The extent of the rejection increased as the MWCO of the membranes decreased, with all of the organic material being rejected by the membrane with a MWCO of 300 Da. Again, this was expected, because most of the constituents of SRNOM have a molecular weight in between 300 Da and 8000 Da [49,50]. Note that NOM in most surface waters consists mainly of humic substances (i.e., lower molecular weight NOM) [32]. Because effective NOM removal is one of the main motivations behind implementing NF and tight UF membranes in drinking water treatment, membranes with lower MWCOs (i.e.,
